Feedback to third draft of the AI Act’s Code of Practice for General-Purpose AI

The AI Act establishes certain obligations for providers of General-Purpose AI (GP-AI) models and those with systemic risks. To facilitate the application of these rules, which will become applicable as of August 2025, the AI Act started the process of setting up a GP-AI Code of Practice. 

Since September 2024, BEUC has been involved in providing input to this Code of Practice. The third draft of the code was published on 11 March and stakeholders were given until 30 March to provide comments.

BEUC welcomes the openness and constructive approach of the chairs and vice-chairs throughout the process to welcome input from stakeholders. For instance, we were grateful for the organisation of a workshop with civil society organizations on 24 March 2025. However, despite certain improvements to the third draft (e.g. the Model Form has to be kept for a period of ten years after the model has been withdrawn), the code still contains significant shortcomings which will negatively impact consumers when using GP-AI systems. Even if we acknowledge that the code is limited by the AI Act's provisions, the law left ample margin for the code to ensure a higher level of consumer protection than in the current version. 

Building on our responses to previous drafts, and acknowledging the limited scope for the final version to have substantial changes, our feedback to the third version focuses on two main points: transparency and systemic risk taxonomy requirements.